文中提出了一种GMM和SVM混合说话人识别模型,在特征参数域和概率得分域对两种模型进行了融合。混合模型结合了GMM和SVM各自的优势,使SVM的概率输出兼顾各说话人模型内部和模型之间的信息,并有效解决了SVM训练算法复杂,难以处理大量样本的问题。采用TIMIT数据库进行了说话人辨认实验,结果证明本文提出的GMM-SVM模型比传统的GMM模型和SVM模型具有更好的辨识性能。 In this paper, a hybrid speaker recognition model of GMM and SVM is proposed, and the two models are fused in the feature parameter domain and the probability score domain. The hybrid model combines the respective advantages of GMM and SVM so that the probability output of SVM takes into account the information inside and between the speaker models and effectively solves the problem that the SVM training algorithm is complex and it is difficult to process a large number of samples. The TIMIT database was used for speaker recognition experiments. The results show that the proposed GMM-SVM model has better performance than the traditional GMM model and SVM model.